Abstract: In response to the challenges of uncertainty and suddenness in order demands within unmanned warehouses, this study proposes an interval forecasting method based on kernel-free optimal fuzzy margin distribution quantile regression averaging. First, a well-designed fuzzy membership function and fuzzy functional margin are introduced to assign differentiated weights to samples, thereby effectively mitigating the impact of outliers on the model. Second, the concept of optimal margin distribution is incorporated into the framework of kernel-free support vector regression to propose the idea of optimal weighted functional margin distribution. By optimizing the margin distribution of the samples, the robustness and generalization capability of the model are significantly enhanced. Finally, multiple feature selection methods are employed to construct variable subsets for model training. The integration of multiple models is achieved via quantile regression averaging, which not only improves the accuracy and efficiency of interval forecasting but also ensures stability. Extensive numerical experiments based on real-order data from an unmanned warehouse operated by the large-scale e-commerce platform JD demonstrate that the proposed method improves order interval prediction accuracy by an average of at least 16.15% compared to various benchmark and state-of-the-art models, offering scientific support for inventory management and personnel scheduling in complex and dynamic environments.

Key words: order demand forecasting, interval forecasting, unmanned warehouse, support vector regression, quantile regression averaging